Harness achieves zero downtime deployments by employing a strategy that supports multiple active versions of its Manager and Delegates, which ensures compatibility and seamless upgrades without disrupting ongoing customer executions. This approach, similar to but more flexible than Blue-Green deployment, allows for instant rollback and frequent production deployments, with rollbacks utilized twice in the last four weeks. The architecture involves a Manager that orchestrates pipeline and workflow executions, coordinating with Delegates that run on customers' premises, allowing complex workflows without accessing customer resources directly. The system handles challenges such as varying network connectivity and upgrade times for Delegates by ensuring all involved processes are compatible, using version-aware entities and HTTP header-based routing for communication. This method allows Harness to deploy updates daily while maintaining service continuity, with a rollback mechanism that quickly reverts to previous versions if necessary.